Recently, Machine Learning has been increasingly applied in journalism in different aspects like automate fact checking, discover hidden facts, and automate workflows. In this paper, we utilize Machine Learning and Natural Language Processing to develop a binary classification model that detects news agency's linguistic style in Arabic. During training phase, we explored different features, it is found that term frequency-inverse document frequency (TF-IDF) correlates more with agency's ideology. The average accuracy of our model was 90%.